Integrations¶

ELIDA complements your existing AI infrastructure. It doesn't replace your gateway, or SIEM orchestrator — it adds the session governance layer between them.

LiteLLM¶

LiteLLM provides a unified API across 100+ LLM providers. ELIDA sits in front of LiteLLM to add session tracking and policy enforcement.

AI Agent → ELIDA (session/policy) → LiteLLM (provider routing) → OpenAI / Anthropic / etc.

Setup¶

# elida.yaml
backend: "http://localhost:4000"  # LiteLLM proxy

# Start LiteLLM
litellm --model gpt-4 --port 4000

# Start ELIDA in front of it
docker run -p 8080:8080 -p 9090:9090 \
  -e ELIDA_BACKEND=http://host.docker.internal:4000 \
  ghcr.io/zamorofthat/elida:latest

# Point your agent at ELIDA
OPENAI_BASE_URL=http://localhost:8080 your-agent

What each does¶

Capability	LiteLLM	ELIDA
Multi-provider routing	Yes	Yes
Request-level load balancing	Yes	No
Session tracking	No	Yes
Kill switch	No	Yes
Policy enforcement	No	Yes
Risk scoring	No	Yes
CDR audit trail	No	Yes

Portkey¶

Portkey is an AI gateway with logging, caching, and request-level controls. ELIDA adds session-level governance on top.

AI Agent → ELIDA (session/policy) → Portkey (gateway) → Providers

Setup¶

# elida.yaml
backend: "https://api.portkey.ai"

# Point ELIDA at Portkey
docker run -p 8080:8080 -p 9090:9090 \
  -e ELIDA_BACKEND=https://api.portkey.ai \
  ghcr.io/zamorofthat/elida:latest

# Agent talks to ELIDA, which proxies through Portkey
OPENAI_BASE_URL=http://localhost:8080 your-agent

What each does¶

Capability	Portkey	ELIDA
Request logging	Yes	Yes (session-scoped)
Caching	Yes	No
Prompt management	Yes	No
Session tracking	No	Yes
Kill switch	No	Yes
Real-time intervention	No	Yes
Risk ladder	No	Yes
Wire-level inspection	No	Yes

Tailscale Aperture¶

Aperture provides identity-aware access control via Tailscale. ELIDA adds behavioral enforcement on top of identity.

AI Agent → Tailscale → Aperture (identity/authz) → ELIDA (session/policy) → Provider

Setup¶

# elida.yaml — use Tailscale identity as session key
session:
  header: "X-Tailscale-User"
  generate_if_missing: false

Configure Aperture to forward traffic to ELIDA, which then proxies to the AI provider. Every ELIDA session is now tied to a Tailscale identity.

What each does¶

Capability	Aperture	ELIDA
Network identity	Yes (Tailscale)	No
ACLs (who can connect)	Yes	No
Session tracking	No	Yes
What the agent is doing	No	Yes (wire inspection)
Kill switch	No	Yes
CDR with identity	No	Yes (via identity headers)

Combined value¶

Neither tool alone provides both identity and behavioral enforcement:

Aperture knows who the agent is, not what it's doing
ELIDA knows what the agent is doing, not who it is
Together: "agent deploy-bot@prod-machine sent rm -rf / — session killed, CDR with full identity attached"

Oso¶

Oso provides agent authorization and monitoring via Tailscale. ELIDA provides wire-level session governance without requiring agent cooperation.

                  ┌─── Oso (agent self-reports via tailnet)
AI Agent ─────────┤
                  └─── ELIDA (inspects wire traffic) ──→ Provider

Key difference¶

Aspect	Oso	ELIDA
Trust model	Agent self-reports	Zero-trust wire inspection
Infrastructure	Requires Tailscale tailnet	HTTP proxy (no dependencies)
Visibility	What agent says it did	What agent actually did
Intervention	Post-hoc alerts	Real-time kill/block/throttle
Setup	SDK + tailnet	`ANTHROPIC_BASE_URL=http://elida`

ELIDA doesn't require agent cooperation. It sits on the wire — the agent can't lie, omit, or refuse to report. If it makes an API call, ELIDA sees it.

OpenTelemetry / SIEM¶

ELIDA exports structured telemetry via OTLP to any compatible collector or SIEM.

ELIDA → OTEL Collector → Cribl / Datadog / Elastic / Grafana / Splunk

Setup¶

# elida.yaml
telemetry:
  enabled: true
  exporter: "otlp"
  endpoint: "your-collector:4317"
  capture_content: "flagged"  # "none", "flagged", or "all"

Capture modes¶

Mode	What gets sent	Use case
`"none"`	Metadata only (no bodies)	Privacy-first, default
`"flagged"`	Only flagged session content	Security monitoring
`"all"`	Every request/response body	Full compliance audit

Violation fields for SIEM correlation¶

Every violation includes structured fields for automated SIEM rules:

{
  "rule": "prompt_injection_ignore_request",
  "severity": "critical",
  "effective_severity": "warning",
  "source_role": "assistant",
  "message_index": 63,
  "event_category": "prompt_injection",
  "framework_ref": "OWASP-LLM01"
}

Claude Code / Cursor / AI IDEs¶

Any AI-powered development tool that uses the Anthropic or OpenAI API can be routed through ELIDA.

Claude Code¶

ANTHROPIC_BASE_URL=http://localhost:8080 claude

Cursor¶

OPENAI_BASE_URL=http://localhost:8080 cursor

VS Code + Continue¶

Set the base URL in Continue's config to point at ELIDA.

What you get¶

Every coding session tracked with full request/response capture
Policy enforcement on tool calls (block dangerous Bash commands)
Kill switch if an agent starts deleting files
Audit trail of everything the AI did in your codebase